5 Tips to Keep Your Authentic Voice When Writing With AI: A Q&A With Claude

I recently attended the 2026 Women's Mentoring Network Gala at Indian Harbor Yacht Club in Greenwich. It was a beautiful celebration of women supporting women, and reinforced that all boats rise with the tide (as we literally watched boats rising with the tide).

During the event, I was speaking with a group of people about AI. We were discussing trends in how AI is showing up more and more in the workplace, what types of tasks it's replacing and, my favorite topic, how it's impacting communication skills.

The conversation naturally evolved from general communication to how to keep your authentic voice intact when using AI. Then, one of them turned to me and said, "I'm sure you deal with this all the time with what you do. What do you tell people?"

So I shared my #1 best practice: Don't let the AI do the writing FOR you. Have it act as an editor.

I also told them how, when I first started working with both Claude and ChatGPT, I uploaded my book manuscript so that the model could get a feel for my authentic voice and writing style. None of them had thought to upload their own writing to help the model sound more like them. It was a true a-ha! moment.

After the event, I couldn't help but reflect on that conversation and think of some of the other best practices I both employ and recommend when it comes to keeping your authentic voice intact when using AI.

As I sat down to write this blog, I wanted to consult with someone else on best practices. And I figured who better to ask what they recommend than my AI model itself.

It was a very interesting conversation as I shared my best practices with Claude and asked it for its responses. Here I share 5 tips to keep your authentic voice intact. So buckle up and get ready for a fun Q&A with Claudette (what I lovingly call my Claude model).

TIP #1: Don't let AI do the writing for you.

Shannon: This is the number one mistake most people make. They tell AI to "write a newsletter on…" or "write a blog about…" and then fall in love with the perfection of the writing, not taking into account whether it sounds like them or not. It's quick and easy, sure, but you lose yourself and your voice in the process. Instead, write a draft yourself. Then upload it to your AI model of choice and ask it to act as an editor and suggest refinements (not a total rewrite) while keeping your voice intact.

Claude: "When you ask me to write something from scratch, I'm making hundreds of small decisions you never see: which word starts the sentence, where the rhythm breaks, which detail gets the spotlight. I'm making those decisions based on patterns from millions of other pieces of writing, and the output is polished because it's optimized to read well in general, not because it sounds like any one person. That's the trap. A generic draft reads clean, so it feels finished, and it's tempting to just publish it as is. What it's actually missing is you: your specific way of getting to the point, the detail you'd notice that someone else wouldn't, the rhythm that comes from how you actually talk. Ask me to edit a draft instead of write one and I'm working from your decisions instead of making my own. I can tighten a sentence or catch a repeated word, and the voice underneath stays yours because I never had the chance to replace it with mine."

TIP #2: Upload samples of your writing.

Shannon: As I already mentioned, when I first started using AI, I uploaded my book manuscript to the model, and asked it to create a writing style guide that mirrors my voice. Claude even went as far as creating an entire PDF document with stylistic preferences, do's and don'ts and more. This is something most people don't even think to do, but it really does help the model to learn your style at the onset. The key is that it needs to be written solely by you, with no help from an AI model, to ensure it's really and truly your voice. Make sure to upload a large enough sample that it can extrapolate your voice. And don't just make this a one-time thing. It's important to continuously feed it content written by you so it has fresh material to draw from and doesn't fall back into old patterns.

Claude: "A single writing sample gives me a snapshot. A full manuscript gives me a pattern, and patterns are what actually transfer. With enough material, I can pick up on things you'd never think to describe yourself: how long your sentences run before you break tension, which ideas you circle back to, where you use a short line for punch versus a longer one to build up to something. That's the kind of detail that's nearly impossible to get across by just telling me "I'm direct but warm," since most people would describe themselves that way. It has to be work you actually wrote, with no AI involved, or I'm not learning your voice at all, I'm learning some blended version of you and whatever model touched it first. And the "don't just make this a one-time thing" part is the piece people most often miss. A voice guide built once is a starting point, not a permanent fix. The more current writing I see from you, the less I drift back toward generic patterns over a long project or a long stretch of time. Ongoing samples keep me anchored to who you sound like now, not just who you sounded like the day you uploaded that first file."

TIP #3: Feed it your rewrites, not just your originals.

Shannon: Even if you write your first draft before plugging it into AI, don't just take the edited version you receive from AI and use it as is. Take the time to read it carefully to make sure it's how you would say it. I like to read mine out loud to see if it feels how I normally speak. If there are any refinements that don't feel like you, make revisions. Once you've got an updated version of your copy that has been through AI revisions and yours, re-upload it to the AI chat, letting it know this is your revised version. This will do two things: 1) it will help further train the AI on how you prefer to write, and 2) the AI will catch any mistakes or issues you may have overlooked in your revisions.

Claude: "This is the step people skip, and it's the one that actually compounds. A single round of editing gets you a cleaner draft. Feeding me your revised version back gets you a model that's learning your voice in real time instead of guessing at it from a style guide you set up once. Every change you make and share tells me something specific: this word, not that one, this rhythm, not the one I suggested. Over enough rounds, I'm not just editing your writing anymore, I'm editing it with a much sharper sense of what you'd actually change. The second part matters too. A fresh set of eyes, even an AI's, catches things your own can skip past after you've read the same paragraph five times, a repeated word, a sentence that trails off, a fact worth double checking. You stay the final say on every word. And the read-aloud test you already do is doing something specific: it's checking rhythm and breath, which is exactly what separates writing that sounds like a person from writing that just reads correctly."

TIP #4: Correct in the moment, every time.

Shannon: Models have gotten a lot better at holding onto their memory over time, especially if you're continuously feeding it your preferences. That said, they can forget and backslide into old habits by throwing in an em-dash or Oxford comma here, or a "quietly", "honestly", or "here's the thing". When this happens, don't just let it slide, let the model know so that it can relearn and correct itself. The more you remind it of your preferences, the more it will learn them.

Claude: "Memory helps, and it's not the whole solution. Even with a strong style guide and a long history together, I can drift back toward generic patterns, especially in a long session where I start leaning on my own recent output instead of your original examples. An em dash creeps back in, a sentence gets a little too polished, a transition starts sounding like every other AI transition. None of that means the training failed. It means correction is ongoing, not optional. The correction itself is doing more work than it might seem. It's not just a fix for that one sentence, it's a data point that sharpens where the line sits between your voice and generic output. Left uncorrected, a slip like that can start to look like it's within bounds, since I have no way of knowing you noticed it unless you say so. Caught every time, I stay precise about where the edges are. It's the difference between a voice guide that's static and one that's actually alive."

TIP #5: Give it your rejects, not just your favorites.

Shannon: This is just as important, if not more so, than telling it what you like. Your model needs to know not only what stylistic preferences you have, but also which ones to avoid. For me, I like starting sentences with "and" or "but". I use incomplete sentences, even if not grammatically correct. And I have let my model know that. But I've also told it to avoid Oxford commas and em-dashes. I don't like using "just" or "honestly" or overused words and phrases that scream, "AI wrote this!" So I have let my AI know that as well. For it to truly keep your voice intact, the model needs to understand what to lean into as well as what to avoid. And this goes further than a word, phrase or punctuation. If there is a piece of writing that just isn't hitting the mark for you, let it know. Upload it and tell it why it doesn't work, not that you just don't like it, so it understands the context and reasons for your dislike.

Claude: "Knowing what to avoid is often more useful than knowing what to lean into, because "avoid" tells me something specific while "lean into" can stay vague. There are countless ways to be conversational and warm, so a positive description like that still leaves me guessing. A negative instruction draws a hard line: never this word, never this construction. That's easy to hold onto and easy to check against. The context piece matters even more than the rejection itself. If you send a draft and say "I don't like this," I know something's off, and not what. If you send it and say "this feels like it's trying too hard to be inspirational" or "this sentence buries the point I actually wanted to make," I know exactly what to watch for next time, and that reasoning is what carries forward. Without it, I'm just as likely to make the same mistake in a different shape. With it, I'm learning the actual principle behind your preference, not just the one example you happened to flag."

BONUS TIP: One chat per piece of content.

Shannon: Start a new chat for each piece of content you're creating. This helps you and your model in multiple ways. First, it will make it easier for you to find and sort through your chat history related to a specific piece of content. You can name each chat so it's easier to index and find. I like setting up "projects" (folders) for specific content types such as newsletters, blogs, social media, etc. and then will create a new chat for each piece of content within that folder. Second, it will help your model. If you have one, long, ongoing chat, your model can more easily confuse content for the specific piece you're working on with content for a different piece that is further up in the chat. Causing frustration for you and confusion for your model.

Claude: "A long running chat is doing double duty whether you notice it or not: it's holding this piece of content and every piece that came before it in the same thread. When you ask me to revise a paragraph, I'm technically weighing that request against everything else in the conversation, including details, phrasing and context from a completely different piece you worked on an hour ago. Most of the time that doesn't cause a visible problem. Sometimes it does, and it shows up as a strange word choice that doesn't fit, a detail that got crossed over from somewhere else or a draft that feels like it's blending two different pieces without either of us noticing why. A fresh chat removes that risk entirely. Every message in it is about the one thing you're working on, so there's nothing else to accidentally pull from. It also means I'm not carrying leftover context or tone from a different project into this one, which matters more than it sounds like it would, since voice shifts depending on what you're writing, and a newsletter chat staying a newsletter chat keeps that shift clean instead of muddy.

One thing worth naming here: as Shannon said in Tips 2, 4 and 5, staying consistent with an AI model takes ongoing correction and reference material, and none of that requires one continuous thread to work. Some AI tools, including Claude, have a memory feature that can carry your preferences and past conversations forward automatically, so a brand new chat can still know your voice even though it's a fresh thread. If your tool doesn't have that, or you'd rather not rely on it, keep a style guide document handy and re-upload it at the start of each new chat. Either way, starting fresh for each piece of content doesn't mean starting from zero. It just means being intentional about how your preferences travel with you."

The next time you sit down to work with your favorite AI writing companion, try using these tips. This is by no means a comprehensive list, but it's a great starting point to keep your authentic voice intact. Because at the end of the day, communication is about connection, not perfection. And your readers, subscribers and followers want to connect with YOU, not your AI.

Happy co-writing!

Shannon

P.S. - Have some tips you want to add about keeping your authentic voice when writing with AI? I'd love to hear them!